Meet Dolly 2.0: Open-source AI for commercial use
Databricks recently released Dolly 2.0, which is an open-source AI that has many similarities with ChatGPT and it is for commercial use.
After releasing Dolly, Databricks didn't waste any time revealing its successor, Dolly 2.0. Even though it is significantly lighter than other known LLMS like GPT3, Dolly 2.0 has doubled its predecessor with 12 billion parameters. Additionally, it has been improved using a high-quality instruction-following dataset gathered from Databricks staff members.
OpenAI has started a new era and reached a point that is very hard for other companies to even get close to. GPT 3 is known for being trained on 175 billion parameters, which is actually insane. However, Dolly 2.0 also has a reasonable capacity and capability when it comes to giving answers to many different questions. Because it is significantly lighter than OpenAI's language model, Dolly can be run on an organization's internal servers without the requirement to exchange data with a third-party company or application.
"We believe models like Dolly will help democratize LLMs, transforming them from something very few companies can afford into a commodity every company can own and customize to improve their products," said Databricks.
Dolly 2.0 is completely open-source
Databricks has made Dolly 2.0 completely open-source, including its training code. The company has also made its dataset available for the public, which has 15,000 human-generated prompts and responses for anyone's use.
The previous model, Dolly 1, didn't let people use it for commercial purposes. However, the company has eliminated the issue with the second model, and now researchers and developers can download it from the official site and access the dataset on GitHub.
"Unlike previous models, Dolly 2.0 provides a license that permits commercial use. That means enterprises can tune, develop, and use Dolly 2.0 using their proprietary datasets. They don't have to give those datasets to a third-party vendor, and they can monetize the model in a way that otherwise is prohibited by existing models. Of course, we hope people will train their models on Databricks, but these models and datasets can be used elsewhere," said Ali Ghodsi, chief executive and co-founder of Databricks.Advertisement